Online Bayesian Learning and Inference for OTHR Target Tracking and Registration
Hua Lan, Yuxiang Mao, Zengfu Wang · IEEE Transactions on Signal Processing · 2024
Coordinate registration (CR), using ionospheric information to map the measurement in radar slant coordinates into geodetic inertial coordinates, plays a crucial role in target tracking of over-the-horizon radar (OTHR). Due to the ionospheric inherent variability and inaccurate modeling, there exists uncertainty in the CR process, decreasing target tracking accuracy. By formulating the OTHR target tracking with uncertain CR as the variational optimization problem, this paper proposes an online Bayesian learning and inference (OBLI) scheme for joint OTHR target tracking and CR. For Bayesian learning, the Gaussian process (GP) models the spatial correlation of the ionosphere with GP hyperparameters learned by streaming sparse GP approximation, which updates the GP hyperparameters and optimizes pseudo-input locations in an online fashion. For Bayesian inference, the streaming variational Monte Carlo approximates the joint posterior distributions of the target state and CR parameters, enabling flexible and accurate nonlinear filtering for non-conjugate models. Bayesian learning enhances CR parameter identification by modeling ionospheric spatial correlation and utilizing prior information. This improvement benefits target tracking by incorporating a joint optimization mechanism of the Bayesian inference. Meanwhile, the proposed OBLI carries out the joint Bayesian learning and inference online, allowing real-time OTHR target tracking applications. Finally, the effectiveness of the OBLI method is verified on OTHR target tracking with uncertain ionospheric heights.